Predicting Takeover Targets - A Machine Learning Approach

Sammanfattning: The possibility of predicting takeover announcements for investment purposes has been widely researched, with varying results and conclusions. This paper builds on previously conducted studies and applies a machine learning approach to the research problem, through the random forest classi er by Breiman (2001). The classi er is known to perform well on imbalanced datasets, which is a key characteristic of takeover data. Furthermore, this paper applies a robust research methodology by extending the time period studied and replacing the choice of an optimal classi cation cuto probability with a range of likelihoods. With this range of probabilities, it investigates the trade-o between predictive accuracy and size of portfolios. We conclude that the addition of the random forest classi er does not bring statistically signi cant superior results compared to previous models. However, our ndings show that it may indeed be possible to earn statistically signi cant abnormal returns through an investment strategy of takeover target prediction. As these ndings are not consistent over time, further research is warranted.